xref: /spdk/module/bdev/raid/concat.c (revision 45a053c5777494f4e8ce4bc1191c9de3920377f7)
1 /*   SPDX-License-Identifier: BSD-3-Clause
2  *   Copyright (C) 2022 Intel Corporation.
3  *   Copyright (c) Peng Yu yupeng0921@gmail.com.
4  *   All rights reserved.
5  */
6 
7 #include "bdev_raid.h"
8 
9 #include "spdk/env.h"
10 #include "spdk/thread.h"
11 #include "spdk/string.h"
12 #include "spdk/util.h"
13 
14 #include "spdk/log.h"
15 
16 struct concat_block_range {
17 	uint64_t start;
18 	uint64_t length;
19 };
20 
21 /*
22  * brief:
23  * concat_bdev_io_completion function is called by lower layers to notify raid
24  * module that particular bdev_io is completed.
25  * params:
26  * bdev_io - pointer to bdev io submitted to lower layers, like child io
27  * success - bdev_io status
28  * cb_arg - function callback context (parent raid_bdev_io)
29  * returns:
30  * none
31  */
32 static void
33 concat_bdev_io_completion(struct spdk_bdev_io *bdev_io, bool success, void *cb_arg)
34 {
35 	struct raid_bdev_io *raid_io = cb_arg;
36 
37 	spdk_bdev_free_io(bdev_io);
38 
39 	if (success) {
40 		raid_bdev_io_complete(raid_io, SPDK_BDEV_IO_STATUS_SUCCESS);
41 	} else {
42 		raid_bdev_io_complete(raid_io, SPDK_BDEV_IO_STATUS_FAILED);
43 	}
44 }
45 
46 static void concat_submit_rw_request(struct raid_bdev_io *raid_io);
47 
48 static void
49 _concat_submit_rw_request(void *_raid_io)
50 {
51 	struct raid_bdev_io *raid_io = _raid_io;
52 
53 	concat_submit_rw_request(raid_io);
54 }
55 
56 /*
57  * brief:
58  * concat_submit_rw_request function is used to submit I/O to the correct
59  * member disk for concat bdevs.
60  * params:
61  * raid_io
62  * returns:
63  * none
64  */
65 static void
66 concat_submit_rw_request(struct raid_bdev_io *raid_io)
67 {
68 	struct raid_bdev_io_channel	*raid_ch = raid_io->raid_ch;
69 	struct raid_bdev		*raid_bdev = raid_io->raid_bdev;
70 	struct concat_block_range	*block_range = raid_bdev->module_private;
71 	uint64_t			pd_lba;
72 	uint64_t			pd_blocks;
73 	int				pd_idx;
74 	int				ret = 0;
75 	struct raid_base_bdev_info	*base_info;
76 	struct spdk_io_channel		*base_ch;
77 	struct spdk_bdev_ext_io_opts	io_opts = {};
78 	int i;
79 
80 	pd_idx = -1;
81 	for (i = 0; i < raid_bdev->num_base_bdevs; i++) {
82 		if (block_range[i].start > raid_io->offset_blocks) {
83 			break;
84 		}
85 		pd_idx = i;
86 	}
87 	assert(pd_idx >= 0);
88 	assert(raid_io->offset_blocks >= block_range[pd_idx].start);
89 	pd_lba = raid_io->offset_blocks - block_range[pd_idx].start;
90 	pd_blocks = raid_io->num_blocks;
91 	base_info = &raid_bdev->base_bdev_info[pd_idx];
92 	if (base_info->desc == NULL) {
93 		SPDK_ERRLOG("base bdev desc null for pd_idx %u\n", pd_idx);
94 		assert(0);
95 	}
96 
97 	/*
98 	 * Submit child io to bdev layer with using base bdev descriptors, base
99 	 * bdev lba, base bdev child io length in blocks, buffer, completion
100 	 * function and function callback context
101 	 */
102 	assert(raid_ch != NULL);
103 	base_ch = raid_bdev_channel_get_base_channel(raid_ch, pd_idx);
104 
105 	io_opts.size = sizeof(io_opts);
106 	io_opts.memory_domain = raid_io->memory_domain;
107 	io_opts.memory_domain_ctx = raid_io->memory_domain_ctx;
108 	io_opts.metadata = raid_io->md_buf;
109 
110 	if (raid_io->type == SPDK_BDEV_IO_TYPE_READ) {
111 		ret = raid_bdev_readv_blocks_ext(base_info, base_ch,
112 						 raid_io->iovs, raid_io->iovcnt,
113 						 pd_lba, pd_blocks, concat_bdev_io_completion,
114 						 raid_io, &io_opts);
115 	} else if (raid_io->type == SPDK_BDEV_IO_TYPE_WRITE) {
116 		ret = raid_bdev_writev_blocks_ext(base_info, base_ch,
117 						  raid_io->iovs, raid_io->iovcnt,
118 						  pd_lba, pd_blocks, concat_bdev_io_completion,
119 						  raid_io, &io_opts);
120 	} else {
121 		SPDK_ERRLOG("Recvd not supported io type %u\n", raid_io->type);
122 		assert(0);
123 	}
124 
125 	if (ret == -ENOMEM) {
126 		raid_bdev_queue_io_wait(raid_io, spdk_bdev_desc_get_bdev(base_info->desc),
127 					base_ch, _concat_submit_rw_request);
128 	} else if (ret != 0) {
129 		SPDK_ERRLOG("bdev io submit error not due to ENOMEM, it should not happen\n");
130 		assert(false);
131 		raid_bdev_io_complete(raid_io, SPDK_BDEV_IO_STATUS_FAILED);
132 	}
133 }
134 
135 static void concat_submit_null_payload_request(struct raid_bdev_io *raid_io);
136 
137 static void
138 _concat_submit_null_payload_request(void *_raid_io)
139 {
140 	struct raid_bdev_io *raid_io = _raid_io;
141 
142 	concat_submit_null_payload_request(raid_io);
143 }
144 
145 static void
146 concat_base_io_complete(struct spdk_bdev_io *bdev_io, bool success, void *cb_arg)
147 {
148 	struct raid_bdev_io *raid_io = cb_arg;
149 
150 	raid_bdev_io_complete_part(raid_io, 1, success ?
151 				   SPDK_BDEV_IO_STATUS_SUCCESS :
152 				   SPDK_BDEV_IO_STATUS_FAILED);
153 
154 	spdk_bdev_free_io(bdev_io);
155 }
156 
157 /*
158  * brief:
159  * concat_submit_null_payload_request function submits the next batch of
160  * io requests with range but without payload, like FLUSH and UNMAP, to member disks;
161  * it will submit as many as possible unless one base io request fails with -ENOMEM,
162  * in which case it will queue itself for later submission.
163  * params:
164  * bdev_io - pointer to parent bdev_io on raid bdev device
165  * returns:
166  * none
167  */
168 static void
169 concat_submit_null_payload_request(struct raid_bdev_io *raid_io)
170 {
171 	struct raid_bdev		*raid_bdev;
172 	int				ret;
173 	struct raid_base_bdev_info	*base_info;
174 	struct spdk_io_channel		*base_ch;
175 	uint64_t			pd_lba;
176 	uint64_t			pd_blocks;
177 	uint64_t			offset_blocks;
178 	uint64_t			num_blocks;
179 	struct concat_block_range	*block_range;
180 	int				i, start_idx, stop_idx;
181 
182 	raid_bdev = raid_io->raid_bdev;
183 	block_range = raid_bdev->module_private;
184 
185 	offset_blocks = raid_io->offset_blocks;
186 	num_blocks = raid_io->num_blocks;
187 	start_idx = -1;
188 	stop_idx = -1;
189 	/*
190 	 * Go through all base bdevs, find the first bdev and the last bdev
191 	 */
192 	for (i = 0; i < raid_bdev->num_base_bdevs; i++) {
193 		/* skip the bdevs before the offset_blocks */
194 		if (offset_blocks >= block_range[i].start + block_range[i].length) {
195 			continue;
196 		}
197 		if (start_idx == -1) {
198 			start_idx = i;
199 		} else {
200 			/*
201 			 * The offset_blocks might be at the middle of the first bdev.
202 			 * Besides the first bdev, the offset_blocks should be always
203 			 * at the start of the bdev.
204 			 */
205 			assert(offset_blocks == block_range[i].start);
206 		}
207 		pd_lba = offset_blocks - block_range[i].start;
208 		pd_blocks = spdk_min(num_blocks, block_range[i].length - pd_lba);
209 		offset_blocks += pd_blocks;
210 		num_blocks -= pd_blocks;
211 		if (num_blocks == 0) {
212 			stop_idx = i;
213 			break;
214 		}
215 	}
216 	assert(start_idx >= 0);
217 	assert(stop_idx >= 0);
218 
219 	if (raid_io->base_bdev_io_remaining == 0) {
220 		raid_io->base_bdev_io_remaining = stop_idx - start_idx + 1;
221 	}
222 	offset_blocks = raid_io->offset_blocks;
223 	num_blocks = raid_io->num_blocks;
224 	for (i = start_idx; i <= stop_idx; i++) {
225 		assert(offset_blocks >= block_range[i].start);
226 		assert(offset_blocks < block_range[i].start + block_range[i].length);
227 		pd_lba = offset_blocks -  block_range[i].start;
228 		pd_blocks = spdk_min(num_blocks, block_range[i].length - pd_lba);
229 		offset_blocks += pd_blocks;
230 		num_blocks -= pd_blocks;
231 		/*
232 		 * Skip the IOs we have submitted
233 		 */
234 		if (i < start_idx + raid_io->base_bdev_io_submitted) {
235 			continue;
236 		}
237 		base_info = &raid_bdev->base_bdev_info[i];
238 		base_ch = raid_bdev_channel_get_base_channel(raid_io->raid_ch, i);
239 		switch (raid_io->type) {
240 		case SPDK_BDEV_IO_TYPE_UNMAP:
241 			ret = raid_bdev_unmap_blocks(base_info, base_ch,
242 						     pd_lba, pd_blocks,
243 						     concat_base_io_complete, raid_io);
244 			break;
245 		case SPDK_BDEV_IO_TYPE_FLUSH:
246 			ret = raid_bdev_flush_blocks(base_info, base_ch,
247 						     pd_lba, pd_blocks,
248 						     concat_base_io_complete, raid_io);
249 			break;
250 		default:
251 			SPDK_ERRLOG("submit request, invalid io type with null payload %u\n", raid_io->type);
252 			assert(false);
253 			ret = -EIO;
254 		}
255 		if (ret == 0) {
256 			raid_io->base_bdev_io_submitted++;
257 		} else if (ret == -ENOMEM) {
258 			raid_bdev_queue_io_wait(raid_io, spdk_bdev_desc_get_bdev(base_info->desc),
259 						base_ch, _concat_submit_null_payload_request);
260 			return;
261 		} else {
262 			SPDK_ERRLOG("bdev io submit error not due to ENOMEM, it should not happen\n");
263 			assert(false);
264 			raid_bdev_io_complete(raid_io, SPDK_BDEV_IO_STATUS_FAILED);
265 			return;
266 		}
267 	}
268 }
269 
270 static int
271 concat_start(struct raid_bdev *raid_bdev)
272 {
273 	uint64_t total_blockcnt = 0;
274 	struct raid_base_bdev_info *base_info;
275 	struct concat_block_range *block_range;
276 
277 	block_range = calloc(raid_bdev->num_base_bdevs, sizeof(struct concat_block_range));
278 	if (!block_range) {
279 		SPDK_ERRLOG("Can not allocate block_range, num_base_bdevs: %u",
280 			    raid_bdev->num_base_bdevs);
281 		return -ENOMEM;
282 	}
283 
284 	int idx = 0;
285 	RAID_FOR_EACH_BASE_BDEV(raid_bdev, base_info) {
286 		uint64_t strip_cnt = base_info->data_size >> raid_bdev->strip_size_shift;
287 		uint64_t pd_block_cnt = strip_cnt << raid_bdev->strip_size_shift;
288 
289 		base_info->data_size = pd_block_cnt;
290 
291 		block_range[idx].start = total_blockcnt;
292 		block_range[idx].length = pd_block_cnt;
293 		total_blockcnt += pd_block_cnt;
294 		idx++;
295 	}
296 
297 	raid_bdev->module_private = block_range;
298 
299 	SPDK_DEBUGLOG(bdev_concat, "total blockcount %" PRIu64 ",  numbasedev %u, strip size shift %u\n",
300 		      total_blockcnt, raid_bdev->num_base_bdevs, raid_bdev->strip_size_shift);
301 	raid_bdev->bdev.blockcnt = total_blockcnt;
302 
303 	raid_bdev->bdev.optimal_io_boundary = raid_bdev->strip_size;
304 	raid_bdev->bdev.split_on_optimal_io_boundary = true;
305 
306 	return 0;
307 }
308 
309 static bool
310 concat_stop(struct raid_bdev *raid_bdev)
311 {
312 	struct concat_block_range *block_range = raid_bdev->module_private;
313 
314 	free(block_range);
315 
316 	return true;
317 }
318 
319 static struct raid_bdev_module g_concat_module = {
320 	.level = CONCAT,
321 	.base_bdevs_min = 1,
322 	.memory_domains_supported = true,
323 	.start = concat_start,
324 	.stop = concat_stop,
325 	.submit_rw_request = concat_submit_rw_request,
326 	.submit_null_payload_request = concat_submit_null_payload_request,
327 };
328 RAID_MODULE_REGISTER(&g_concat_module)
329 
330 SPDK_LOG_REGISTER_COMPONENT(bdev_concat)
331